Happy slice of mcp pizzaMCP.pizza
mcp-bigquery-server

mcp-bigquery-server

MCP.Pizza Chef: ergut

One tool does everything: you ask a question in ordinary words, your assistant writes the query, and only reading is ever permitted. That limit is real rather than a promise, because every query is handed to Google's own planner for inspection first, so anything that would add, change, drop or export data is refused before it runs. Signing in uses your existing Google Cloud account rather than a key, and each query is billed to that account by the amount of data it scans.

Data

Use This MCP server To

Ask who my top ten customers were last month Check a figure before a meeting without writing a query Find out which tables hold the numbers I need Compare this month against the same month last year Pull a quick count without bothering the data team

README

BigQuery MCP Server

BigQuery MCP Server Logo

What is this? šŸ¤”

This is a server that lets your LLMs (like Claude) talk directly to your BigQuery data ā€” read-only, with no ability to mutate your warehouse. Think of it as a friendly translator that sits between your AI assistant and your database, making sure they can chat securely and efficiently.

Quick Example

You: "What were our top 10 customers last month?"
Claude: *queries your BigQuery database and gives you the answer in plain English*

No more writing SQL queries by hand - just chat naturally with your data!

How Does It Work? šŸ› ļø

This server uses the Model Context Protocol (MCP), which is like a universal translator for AI-database communication. MCP is supported by Claude Desktop, Claude Code, and a growing number of other AI clients.

Here's all you need to do:

  1. Set up authentication (see below)
  2. Add your project details to your MCP client's config file
  3. Start chatting with your BigQuery data naturally!

What Can It Do? šŸ“Š

mcp-bigquery-server FAQ

Can it change or delete anything in my warehouse?
No, and this claim holds up when you read the code. Every query is submitted to Google for inspection before it runs and is rejected unless it is purely a read. Inserts, updates, deletes, drops, merges and exports are all refused, in every mode.
Does it cost money?
Indirectly, yes. Google charges for the amount of data each query scans, and that lands on your own cloud bill. A cap of one gigabyte per query applies unless you raise it, and the sample settings file raises it to ten, so a chatty assistant can add up.
Do I need a key?
Not a key in the usual sense. You either sign in once with Google's own command-line tool or point it at a service account file. It can see exactly what your own account is allowed to see, and nothing more.
Will it show sensitive columns like emails or dates of birth?
By default, yes. The quick setup everyone follows runs with no column restrictions at all. They only switch on once you add a settings file, which can then block named columns and even scan your tables for likely personal details. This matters because query results are sent onward to whichever model you chat with.
Which apps does it work in?
Claude Desktop and Claude Code are documented, and any other app that can start a local MCP server works the same way. Setup is one sign-in command plus a short block naming your cloud project.
Can I use this to answer a business question without writing any code?
Yes, that is the entire point. You ask in plain words, the assistant works out the query, and you get an answer back in plain words rather than a grid of raw rows.
How many tools does it add?
Just one, for running a read query. It also lets your assistant browse your datasets and see what columns each table has, which is what makes the plain-English questions work.
Is it still maintained?
Yes. The most recent code change was May 2026, the protections have a substantial test suite behind them, and the licence is the permissive kind with no unusual conditions.